A STUDY OF ERRORS IN VISUAL SAMPLING USING THE THEORY OF SAMPLING (TOS)
Mohammadamin Ashkbari, Jean-Sébastien Dubé, François Duhaime
In the proceedings of: GeoManitoba 2025: 78th Canadian Geotechnical Conference & 9th Canadian Permafrost ConferenceSession: Soil Mechanics & Foundations 2
ABSTRACT: This study evaluates the performance of a visual sampling method for estimating the particle size distribution (PSD) of granular soils using image-based analysis. A convolutional neural network (CNN) model previously trained for PSD prediction was employed to analyze 32 replicate images under dry conditions and 32 replicate images under moist conditions (6% water content). The results show that moisture addition significantly improves reproducibility by reducing expanded uncertainty bounds across all particle sizes. While moist sampling introduced a slight increase in bias at finer fractions, the variability was considerably reduced. The total relative sampling variance decreased from 0.002610 for dry samples to 0.000753 for moist samples. These findings suggest that moist sampling improves precision by stabilizing surface conditions, despite a slight increase in bias at finer particle sizes.
RÉSUMÉ: Cette étude évalue la performance d'une méthode d'échantillonnage visuel pour estimer la distribution granulométrique (PSD) de sols granulaires par analyse d'images. Un modèle de réseau neuronal convolutif (CNN) préalablement entraîné pour la prédiction du PSD a été utilisé pour analyser 32 images en conditions sèches et 32 images en conditions humides (6 % d'humidité). Les résultats montrent que l'ajout d'humidité améliore considérablement la reproductibilité en réduisant les incertitudes élargies pour toutes les tailles de particules. Bien que l'humidité introduise une légère augmentation du biais systématique pour les particules fines, la variabilité globale est nettement réduite. La variance relative totale passe de 0,002610 pour les échantillons secs à 0,000753 pour les échantillons humides. Ces résultats démontrent que l’humidification avant l’échantillonnage visuel stabilise la surface et améliore la précision des estimations de la distribution granulométrique, bien que la représentativité globale puisse être réduite en raison d’un biais accru pour les fractions fines.
Submit a Data Update Form for this paper
Please include this code when submitting a data update via other methods: GEO2025_282
Access this article:
Canadian Geotechnical Society members can access to this article, along with all other Canadian Geotechnical Conference proceedings, in the Member Area. Conference proceedings are also available in many libraries.
Cite this article:
Ashkbari, Mohammadamin, Dubé, Jean-Sébastien, Duhaime, François (2025) A STUDY OF ERRORS IN VISUAL SAMPLING USING THE THEORY OF SAMPLING (TOS) in GEO2025. Ottawa, Ontario: Canadian Geotechnical Society.
@inproceedings{Ashkbari_GEO2025_282,
author = {{Ashkbari, Mohammadamin}, {Dubé, Jean-Sébastien}, {Duhaime, François}}
title = {A STUDY OF ERRORS IN VISUAL SAMPLING USING THE THEORY OF SAMPLING (TOS) }
booktitle = {Proceedings of the 78th Canadian Geotechnical Conference & 9th Canadian Permafrost Conference}
year = {2025}
organization = {The Canadian Geotechnical Society},
address = {Ottawa, Canada} }
title = {A STUDY OF ERRORS IN VISUAL SAMPLING USING THE THEORY OF SAMPLING (TOS) }
booktitle = {Proceedings of the 78th Canadian Geotechnical Conference & 9th Canadian Permafrost Conference}
year = {2025}
organization = {The Canadian Geotechnical Society},
address = {Ottawa, Canada} }
